diff --git a/lerobot/common/robot_devices/control_utils.py b/lerobot/common/robot_devices/control_utils.py
index 76e264999..68f69487b 100644
--- a/lerobot/common/robot_devices/control_utils.py
+++ b/lerobot/common/robot_devices/control_utils.py
@@ -243,6 +243,11 @@ def control_loop(
 
     timestamp = 0
     start_episode_t = time.perf_counter()
+
+    # Controls starts, if policy is given it needs cleaning up
+    if policy is not None:
+        policy.reset()
+
     while timestamp < control_time_s:
         start_loop_t = time.perf_counter()